Re: [PATCH v5 0/3] MR7237: include: Fix ReadNoFence64 on i386.

[William Horvath (＠whrvt)]

William Horvath (@whrvt) commented about include/winnt.h:
> #if defined(__x86_64__) || defined(__i386__)

```suggestion:-0+0
#if !defined(__clang__) && (defined(__GNUC__) && (__GNUC__ < 8 || (__GNUC__ == 8 && __GNUC_MINOR__ < 1)) && (defined(__x86_64__) || defined(__i386__)))
```

since this was [fixed](https://gcc.gnu.org/git/?p=gcc.git&a=commit;h=d8c40eff56f69877b33c697ded...) quite a while ago? This would make the include guards that your patch adds elsewhere quite messy, but I think the intrinsics should be preferred if they're not broken. Is there another reason we avoid them completely, besides what's in the GCC bugzilla link?